1 Using Webcrawlers to Estimate Incremental Measure Costs for the Retail Plug-Load Portfolio (RPP) Program November 21, 2014.

1

Using Webcrawlers to Estimate Incremental Measure Costs for the Retail Plug-Load Portfolio (RPP) Program

November 21, 2014

Webcrawlers and RPP

• Agenda:• Review of webcrawling as applied to LED lamps• Applying webcrawlers to other new product categories• Comparison of the webcrawling methodology to the CPUC’s IMC study

2

Applications of Webcrawling and Big Data to LED Lamps

• Project Goal: Understand relationship between price and performance for LED lamps. • Study metrics: Light color, light quality, efficacy, lifetime, and dimmability.

• Research Implications: • Identify changes in costs over time• Look beyond efficiency toward product performance

3

2012 Analysis Conducted by Energy Solutions for PG&E

• Approach:• 700 unique price points were manually collected for over 500

unique lamp models • Multi-variable regression model to analyze the dataset

4

ENERGY STAR?

CRI

CCT

Power Factor

Wattage

Efficacy

Light Output

Bulb Shape

Dimmability Lifetime

Price Modeling – 2012 Data

5

Note: Results based on online retailer data, which we found to be significantly higher on average than in store prices.

2014 Analysis: Applying Big Data

• Retailer-based web crawler tool:• Retailer provided APIs (Application Programming Interfaces)• Screen-scraping methods

• Scope of data collection:• 9 online retailers• 3,000 unique price points• 1,000 unique LED lamp models• 50 different manufacturers• Data collected weekly

6

A Difference in Magnitude: 2012 vs. 2014 Data Collection

7

Watts

Lamp Sh

ape / Typ

e

Lumens

Lumen M

aintenance

Color Tempre

rature

Efficacy

(lpw)

Dimmable (Y

/N)

Energy S

tar Qual. (

Y/N)

Warra

nty CRI

Power Facto

r

Beam Angle

Input Volta

ge R9

Color Consis

tency /

Change

Candlepower (intensit

y)

Product

Weight

Power Typ

e (AC/D

C)0

500

1000

1500

2000

2012 Data - Manually Collected 2014 Data - Web-Crawler

Lamp Property

# Pr

oduc

ts

Note: 2014 data is refreshed every week

Some Benefits of Big Data

• More data -> improvements to the regression analysis:• Individual models could be created for each lamp type• Additional independent variables analyzed• Comparable or improved explanatory power for each model

• New data is collected each week with minimal effort• Ability to monitor real-time performance and price changes• Observe trends in performance and price

8

Sample Regression Results

Best fit model is based on:

• Lumens • Brand• Energy Star Qualified

Metrics not independently impacting price include:

• Dimmable• Color Temperature• CRI• Wattage• Beam Angle• Warranty Length• Diameter• Efficacy• Lumen Maintenance

9

Observed Trends

10

Implications for IMC over Time

11

No more IMC for CRI?

• Moving towards a more dynamic understanding of IMC and performance

Developing a Dynamic Understanding of Products

• IMC is one of many issues that can be informed by data analysis

What’s the market’s baseline performance?

How do the best products perform?

How is performance changing over time?

What’s the incremental cost?

12

Outstanding Questions for this Effort

13

• Opportunities for Refinement• Identifying new ways to use data most effectively• Linking to product performance databases• Inconsistent retailer info and labeling• Developing improved links between in-store and online data• Better understanding regional pricing impacts (rebates)

• Remaining Questions• Legal issues of web-crawling

Applications of Webcrawlers and Big Data

• Agenda:• Review of webcrawlers applied to LED lamps• Applying webcrawlers to new product categories• Comparison of the webcrawling methodology to the CPUC’s IMC study

14

Building on Prior Work

• Ongoing Parallel Efforts• LED bulbs webcrawlers• Assess compliances with CEC Appliance Standards• Development of paper on scaling webcrawler effort

• Existing Infrastructure• Established extraction mechanisms for 9 different online retailers (Best

Buy, Costco, Lowe’s, Home Depot, Walmart, ACE Hardware)

15

Applying Past Experience to New Product Categories

• For New Products • Straightforward for retailers in previous effort• Identify key product features to track• Perform initial data requests• Refine process over time to improve data quality

• For New Retailers • Requires significantly more time to understand retailer API, website

layout.

16

Sample Product Features: Refrigerators

• Primary Product Features:• Brand• Model• Configuration• Defrost Type• Through the Door Ice, Water• Total Capacity (Freezer, Fresh)

• Energy Data

• Energy Star Qualified?• Annual Electricity Use

17

Applications of Webcrawlers and Big Data

• Agenda:• Review of webcrawlers applied to LED lamps• Applying webcrawlers to new product categories• Comparison of the webcrawling methodology to the CPUC’s IMC study

18

IMC Estimation: A Comparison of Methodologies

• Traditional IMC Studies:• Given high cost and time to

execute, restricted to high priority measures

• Managed by the CPUC• Take 2-3 years to complete• Conducted every 3-5 years

• Webcrawler Approach:• Better suited to rapidly-

changing markets• Use a similar analytic approach

(e.g., hedonic price models)• Census of products on all

available site• New products can be added as

needed• Data collected faster and in

higher volume• Data must be adjusted for

differences between brick-and-mortar and online price points

19

Next Steps

• Test approach on 1-2 products, expand to remaining products

• Timeline:• May be between 1-2 months to develop IMC estimates for all product

categories for current retailers.• Additional time is required for additional retailers

20

Questions?